How MIT’s Project NANDA Aims To Decentralize AI Agents

MIT's Project NANDA is building open, decentralized infrastructure for AI agents as an alternative to proprietary platforms like OpenAI's ChatGPT. The project features four architectural layers: a globally distributed Index (DNS for agents), AgentFacts for identity verification, a federated Registry Quilt, and cross-protocol support for MCP, A2A, and NLWeb. NANDA aims to enable agent discovery, authentication, and interaction without central control, similar to how ActivityPub powers the fediverse. While the project has launched developer tools like Join39 and List39 with over 1,000 registered agents, its future adoption remains uncertain as it competes with emerging proprietary agent platforms.